METHODS

published: 03 June 2015

doi: 10.3389/fnins.2015.00202

Optimal control based seizure

abatement using patient derived
connectivity
Peter N. Taylor 1*, Jijju Thomas 2 , Nishant Sinha 3 , Justin Dauwels 3 , Marcus Kaiser 1, 4 ,
Thomas Thesen 5 and Justin Ruths 2
1
Interdisciplinary Computing and Complex BioSystems (ICOS) Research Group, School of Computing Science, Newcastle
University, Newcastle upon Tyne, UK, 2 Engineering Systems and Design, Singapore University of Technology and Design,
Singapore, Singapore, 3 School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore,
Singapore, 4 Institute of Neuroscience, Newcastle University, Newcastle upon Tyne, UK, 5 Department of Neurology, New
York University, New York, NY, USA

Epilepsy is a neurological disorder in which patients have recurrent seizures. Seizures

occur in conjunction with abnormal electrical brain activity which can be recorded
by the electroencephalogram (EEG). Often, this abnormal brain activity consists of
high amplitude regular spike-wave oscillations as opposed to low amplitude irregular
oscillations in the non-seizure state. Active brain stimulation has been proposed as a
Edited by:
ShiNung Ching,
method to terminate seizures prematurely, however, a general and widely-applicable
Washington University in St. Louis, approach to optimal stimulation protocols is still lacking. In this study we use a
USA
computational model of epileptic spike-wave dynamics to evaluate the effectiveness
Reviewed by:
of a pseudospectral method to simulated seizure abatement. We incorporate brain
Mark A. Kramer,
Boston University, USA connectivity derived from magnetic resonance imaging of a subject with idiopathic
Ishita Basu, generalized epilepsy. We find that the pseudospectral method can successfully generate
Massachusetts General Hospital, USA
time-varying stimuli that abate simulated seizures, even when including heterogeneous
*Correspondence:
Peter N. Taylor,
patient specific brain connectivity. The strength of the stimulus required varies in different
School of Computing Science, brain areas. Our results suggest that seizure abatement, modeled as an optimal control
Newcastle University, Newcastle upon
problem and solved with the pseudospectral method, offers an attractive approach to
Tyne NE1 7RU, UK
[email protected] treatment for in vivo stimulation techniques. Further, if optimal brain stimulation protocols
are to be experimentally successful, then the heterogeneity of cortical connectivity should
Specialty section: be accounted for in the development of those protocols and thus more spatially localized
This article was submitted to
Neuroprosthetics, solutions may be preferable.
a section of the journal
Keywords: optimal control, numerical methods, epilepsy model, connectome, bistability, spike-wave, stimulation
Frontiers in Neuroscience

Received: 03 April 2015

Accepted: 21 May 2015
Published: 03 June 2015
1. Introduction
Citation: Epilepsy is a spectrum of disorders in which patients have seizures due to abnormal neuronal
Taylor PN, Thomas J, Sinha N, activity (Blumenfeld and Taylor, 2003). Symptomatic manifestations of these events can include
Dauwels J, Kaiser M, Thesen T and
a loss of consciousness, tonic-clonic convulsions and myoclonic jerks, amongst others which
Ruths J (2015) Optimal control based
seizure abatement using patient
can severely impact patient quality of life. These transient seizure events often have distinctive
derived connectivity. electrographic correlates detectable on the electroencephalogram (EEG). One commonly observed
Front. Neurosci. 9:202. electrographic seizure manifestation is the spike wave discharge (SWD). SWDs are high amplitude
doi: 10.3389/fnins.2015.00202 periodic oscillations with a frequency typically slower then that of normal awake EEG. They are

Frontiers in Neuroscience | www.frontiersin.org 1 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

often associated with absence seizures, myoclonic seizures and advantages in terms of being less invasive and more robust over
complex partial seizures (Asconapé and Penry, 1984; Sadleir et al., alternative methods that employ feedback. We test the robustness
2006). Currently the first line of treatment for patients with of our method by applying the approach in different settings.
epilepsy is typically medication, however in over 30% of cases We begin with a relatively simple model which neglects spatial
medication alone is insufficient (Keränen et al., 1988). interactions and ultimately build up to large-scale control of
Brain stimulation has been suggested as an alternative a stochastic model using connectivity derived from a patient
therapeutic treatment for epilepsy (Liang et al., 2010; Berényi with clinically diagnosed idiopathic generalized epilepsy. To our
et al., 2012; Liang et al., 2012; Saillet et al., 2012). In knowledge this is the first epilepsy modeling study using patient
addition, it has also been suggested that noninvasive stimuli derived diffusion MRI based connectivity and consequently also
such as an auditory tone (Rajna and Lona, 1989) or through the first attempt to control seizures in such a model.
the use of transcranial magnetic stimulation (TMS) (Conte
et al., 2007) could be used to interrupt SWD seizures in 2. Materials and Methods
humans. Unfortunately optimal parameters for stimulation
for the abatement of SWD seizures are currently unknown. 2.1. Imaging
Attempting to elucidate optimal control parameters in an Cortical connectivity was inferred from a 22 year old female
experimental/clinical setup is difficult due to various ethical, patient clinically diagnosed with idiopathic generalized epilepsy
safety and financial reasons. with a history of absence and generalized tonic clonic seizures.
In silico testing of stimulation protocols offers a The subject gave their written informed consent to participate in
complementary approach to in vivo experimentation. Indeed, this study, which was approved by the Institutional Review Board
several computational models of epileptiform SWD exist at the of NYU Langone School of Medicine. T1 structural MRI and
macroscopic spatial scale which is routinely recorded clinically DTI images were acquired using a Siemens Allegra 3T scanner.
using EEG. However, many of these models treat the cortex as Diffusion images were collected using 64 directions, with a b-
a spatially continuous homogeneous medium (Robinson et al., factor of 1000 s mm−2 , one b0 image and 2.5 mm isovoxel, TR =
2002; Breakspear et al., 2006; Marten et al., 2009), or disregard 5500 ms, TE = 86 ms. A T1 anatomical image also acquired using
spatial interactions altogether (Wang et al., 2012). In contrast, it the following parameters: TR = 2530 ms, TE = 3.25 ms, FOV =
has been suggested that spatial heterogeneities may be important 256 mm at a resolution of 1 × 1 × 1.33 mm.
in seizure genesis or maintenance (Westmijse et al., 2009; To infer the cortico-cortical connectivity of the patient we
Kramer and Cash, 2012; Terry et al., 2012) and should therefore first, using the T1 image, segmented white matter and gray matter
be incorporated into a model (Baier et al., 2012). areas, then performed parcellation of the gray matter into 66
Recent years have seen the development of brain imaging regions of interest. These regions of interest correspond to major
protocols using magnetic resonance imaging (MRI) which gyral-based anatomical areas which have been shown to be highly
enable the inference of heterogeneous subject-specific brain consistent between subjects (Desikan et al., 2006). These gray
connectivity. It is essentially possible to generate a connectivity matter volume ROIs generated using FreeSurfer (http://surfer.
matrix representing the brain network, with brain areas nmr.mgh.harvard.edu) were then imported into DSI studio
represented by nodes, and edges/connections inferred using (Yeh et al., 2010) along with the motion corrected diffusion
tractography algorithms passing through the white matter. The images. Whole brain seeding was then used and tractography
so-called structural connectome (Sporns et al., 2005), represented was performed. Only tracts with both ends terminating in the
as a matrix, can be directly incorporated into a computational gray matter were retained. When a total of 5,000,000 tracts were
model of brain activity. Several previous studies have used this found tractography was terminated. With the tracts and the ROIs
approach to simulate healthy brain function (Honey et al., 2009; registered to the same space the mean fractional anisotropy along
Deco et al., 2013; Haimovici et al., 2013; Messé et al., 2014). tracts connecting two ROIs was then taken as a connectivity
However, very few have simulated epileptic activity (Taylor et al., weight. This weighted structural connectivity matrix (M) is then
2013b, 2014a; Yan and Li, 2013). used in the model to directly represent cortical connectivity of the
The control of a system with SWD oscillations is highly patient. Figure 1 summarizes the image processing. A full list of
nontrivial since the system is nonlinear (Taylor et al., 2014b). ROI names can be found in Table S1.
The goal of seizure abatement through stimulation can be cast
as an optimal control problem, which provides a systematic 2.2. Model
and general approach for designing stimuli. Control theory’s 2.2.1. Spatially Independent
traditional analytical techniques, however, do not scale well as the Experimental evidence suggests important roles for both the
size of the system increases, as is the case in considering a model cortex and thalamus in the genesis and maintenance of
with spatial heterogeneities. In recent years the pseudospectral epileptic SWD oscillations (Destexhe, 1998; Pinault and O’Brien,
method has been applied successfully in a variety of applications 2005). We therefore incorporate knowledge of these anatomical
as a highly efficient, robust method for the control of large-scale structures into our model using neural field equations based
nonlinear systems (Ruths and Li, 2012). In this study we use the on the Amari framework (Amari, 1977) which has been
pseudospectral method to design time-varying stimuli for SWD previously used to model SWD (Taylor and Baier, 2011;
seizure abatement in silico cast as optimal control problems. The Taylor et al., 2014b). The cortical subsystem is composed
open-loop controls developed by this technique offer distinct of excitatory pyramidal (PY) and inhibitory interneuron

Frontiers in Neuroscience | www.frontiersin.org 2 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

in which x = PY, IN, TC, RE and ǫ determines the sigmoid

steepness. We simplify the thalamic subsystem by using a linear
activation term s[x] = ax + b instead of the sigmoid function
f [x] since this does not qualitatively impact the dynamics and
makes analysis simpler (Taylor et al., 2014b). This follows the
connection schematic as shown in Figure S1 based on Pinault and
O’Brien (2005).
Deterministic model solutions of Equation (1) are simulated
numerically using ode45 in MATLAB. Stochastic model solutions
are computed numerically using a fixed step Euler-Maruyama
solver in MATLAB with a step size (h) of 1/15000 s. Equations
for the noise driven system are given in Supplementary Methods
Section 1. Stimulations to induce SWD are simulated as a
perturbation to the PY and IN variables in state space where the
control (stimulus) u(t) is applied to the cortical variables only.
Parameters are identical to those used in Taylor et al. (2014b).

2.2.2. Spatially Extended

Following simulations with only one cortical area, the model can
easily be extended to include multiple cortical areas. In our model
the cortical areas have local connectivity within an area through
reciprocal PY→IN and IN⊣PY connections in addition to long
range excitatory connections only. Long range connections (on
the order of several centimeters in length) have been shown
experimentally to be predominantly excitatory. We therefore
incorporate this into our model using the patient-specific DTI
matrix M to represent PY↔PY connections. This approach
of incorporating long range connectivity as excitatory is in
FIGURE 1 | MRI processing and modeling pipeline. A patient-specific agreement with previous modeling studies (Babajani-Feremi and
connectivity matrix is generated using anatomical T1 and diffusion weighted Soltanian-Zadeh, 2010) and follows the connectivity schematic
MRI. Segmentation and parcellation are performed using FreeSurfer (blue
and equation in Supplementary Methods Section 2.
arrow) to define network nodes and tractography is performed using DSI
Studio (red arrows) to determine connections in the network. Custom Matlab
code is used to import the connectivity and simulate the model (orange
arrows). 2.3. Optimal Control
Broadly speaking, optimal control is a mathematical framework
for systematically selecting the time-varying input needed to
drive a dynamical system in a desired way. In general, many
(IN) populations. The thalamic subsystem includes variables choices of input, or stimuli, might achieve a desired objective
representing populations of thalmocortical relay cells (TC) and and without the formalism of optimal control selecting one of
neurons located in the reticular nucleus (RE). All populations these options from a family of potential stimuli is ad-hoc and ill-
are interconnected in agreement with experimentally known defined. An optimal control problem couples a cost, or fitness,
connections (Pinault and O’Brien, 2005) using the connectivity function to be minimized (or potentially maximized) with a set
parameters C1...9 . The resulting model equations are therefore: of constraints. Setting it apart from conventional optimization
˙ problems is that this set of constraints includes the differential
PY(t) =τ1 (hpy − PY + C1 f [PY] (1)
(or difference) equation that captures the dynamics of the system
− C3 f [IN] + C9 f [TC]) + u(t) (Luenberger, 1968). Initial (at the start time, t = 0) conditions
˙
IN(t) =τ2 (hin − IN + C2 f [PY]) + u(t) and often final (at the final time, t = T) constraints also exist.
˙ Path constraints that are imposed over the entire time window
TC(t) =τ3 (htc − TC − C6 s[RE]
t ∈ [0, T] are also possible. Most critically, the cost function
+ C7 f [PY]) must be selected appropriately to evaluate the candidate options
˙
RE(t) =τ4 (hre − RE − C4 s[RE] of stimuli and select the correct one.
+ C5 s[TC] + C8 f [PY]) While the framework of optimal control can capture such a
desired objective well, the techniques to solve optimal control
where hpy,in,tc are input parameters, τ1...4 are timescale problems analytically are limited, especially for large-scale
parameters and f [x] is the sigmoid function : and nonlinear systems. We, therefore, turn to computational
methods to solve them. The pseudospectral method is an ideal
f [x] = (1/(1 + ǫ −x )) (2) computational method for this purpose, namely for practitioners

Frontiers in Neuroscience | www.frontiersin.org 3 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

in a variety of applied disciplines to use, due to its high level of of any lateral spatial interactions (Equation 1). Figure 2A shows
accuracy and ease of implementation. the maxima and minima of the model output for different
The method benefits, like other spectral methods (e.g., values of the parameter htc . For more negative values shown
Fourier series), from the exponential convergence, as the (htc <≈ − 2, left side of figure) there is only one stable solution,
order of approximation increases, characteristic of orthogonal all simulations converge to the steady state (stable focus). For less
functions (Fornberg, 1998). In this case we use the Legendre negative values (−2 <≈ htc < −1.5, shaded area of figure)
polynomials to approximate the states and control. The method a bistable region exists between the stable focus and the SWD
also relies (the “pseudo” part of the name) on a recursive oscillations. This arises following a fold of cycles bifurcation at
relation between the Lagrange interpolating polynomials and htc ≈ −2. Beyond the disappearance of the stable focus (due to
the Legendre polynomials, so that the approximation can be a subcritical Hopf bifurcation) at htc > −1.5, monostable SWD
instead approximated by Lagrange polynomials, leading to and slow waves exist (right hand side of figure). In the bistable
a double approximation: the unknown states/controls to the region a separating manifold (separatrix) exists between the two
Legendre approximation to the Lagrange approximation (Canuto states in four dimensional state space. This manifold is highly
et al., 2006). As the second approximation is an interpolation, complex in structure (Taylor et al., 2014b).
the coefficients of the Lagrange approximation are the values The stable focus can be considered analogous to resting state
of the states and controls themselves at the discretized time background EEG, and the high amplitude oscillatory attractor to
points, rather than more abstract coefficients of the Legendre be the seizure state (Kalitzin et al., 2010; Taylor et al., 2014b).
expansion. The latter case (where abstract coefficients are used) Transitions between non-seizure and seizure states can occur
is what occurs in a Fourier series approximation of a signal. when a stimulus beyond the separatrix occurs. When this does
The coefficients have an interpretation, but the information occur in the bistable region a further stimulus is required to stop
gleaned is indirect information about the signal itself. These the SWD, if a second stimulus is not given the SWD will continue
two factors, the pseudo and spectral, make the method both indefinitely. In Figure 2B we show an example time series
easy to implement, efficient, and, when combined with standard following such a stimulus. In the region immediately preceding
nonlinear optimization solvers, a powerful and scalable approach the bifurcation at htc ≈ −2 complex excitable transients occur
for solving optimal control problems. lasting several seconds (Figure S3A). Ultimately the goal of
Ultimately, the pseudospectral method utilizes these stimulus driven seizure abatement is to minimize the duration
approximations to discretize (in time) the continuous optimal of the seizure following detection.
control problem into a nonlinear optimization problem. The
decision variables of the subsequent optimization problem are 3.2. Optimal Control of Deterministic Spike-Wave
the coefficients of the Lagrange interpolating polynomial, which Dynamics
are also the values of the unknown state and control functions The control of SWD implemented here requires a two-step
at the discretization points. This optimization problem can process; seizure detection and seizure control. The seizure is
be solved using any number of commercial or open-source detected when the PY and IN variables are in the proximity of
nonlinear solvers. While nonlinear optimization is still a field of a point specified on the bistable limit cycle. This could easily be
much research, the work to-date has produced a number efficient
algorithms that scale well on large-scale problems. In order to
recover the state and control functions from the discretized
solution, we construct the Lagrange approximating polynomial
from the optimal decision variables.
A complete presentation of the pseudospectral method and
implementation can be found in the supplementary text.
In this work, we use a cost that minimizes the input power
(the integrated square of the input). Such a cost function both
reduces the invasiveness of the stimuli and also tends to produce
inputs that are more interpretable, as they are devoid of non-
essential fluctuations in the control shape. We also impose state
constraints at the initial and final time to enforce the desired
state transfer. Finally, time is discretized into 81 nodes (using
a Lagrange approximation of 81 terms), which is dramatically
smaller when compared with other methods, such as Runge-
Kutta.

3. Results
3.1. Model Dynamics FIGURE 2 | Bifurcation diagram. (A) Minima and maxima of time series for
different values of htc . A fold of cycles bifurcation occurs at the transition
We begin with the simplest of our scenarios. We investigate the between bistability and excitability. (B) Time series of the model output.
model without noise (i.e., purely deterministic) and independent

Frontiers in Neuroscience | www.frontiersin.org 4 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

adapted in an experimental setting by using delay embedding to shows a clinical recording of one EEG channel from a patient
predict state variables (Takens, 1981; Babloyantz and Destexhe, exhibiting transitions between non-seizure and seizure states.
1986; Taylor et al., 2014b). Since the SWD is fairly regular This compares favorably with the stochastic model simulation
between cycles and between seizures this “trigger point” can be (Figure 4B). Irregular oscillations around the stable focus driven
used, provided that the seizure activity passes close by in state by noise resemble background activity with an abrupt onset
space (e.g., within an error tolerance of ±10%). In theory all of SWD. Figure 4B shows a simulation without any external
points on the SWD limit cycle could be used as trigger points control. In Figure 4C, the same control signal in Figure 3 is
to decrease the time taken to detect and subsequently control the applied.
seizure, where each point would correspond to a stimulus with The challenge with dealing with stochasticity and the success
a different profile. This would mean that the stimulus could be here with this method underscores the importance of a systematic
applied at any phase in the spike. However, we limit ourselves in approach to seizure abatement. Because the optimal control
this study to a single arbitrarily chosen point and leave optimal drives the system from near a known trigger point in state space
seizure detection for future study. to the background state, the effects of stochasticity are minor.
Once the SWD has passed close enough to the trigger point Ad-hoc approaches that work in the deterministic case, may be
the seizure is detected and the control stimulus is applied starting highly sensitive to the perturbations introduced when noise is
at that time instant. Figure 3B shows the state space for the PY added. In previous work (Ruths et al., 2014), we demonstrate how
and IN variables. A stimulus to initiate a seizure is indicated by ensemble control can be used to develop stimuli that are robust
an arrow, while the red △ indicates the trigger point. In both the to variation in the initial state. This situation arises when the
bistable and excitable cases the seizure is abated prematurely by noise driven process and the delays in triggering cause the state
the control (red lines in Figure 3A). An important advantage of to shift noticeably before the control can be applied. Because of
the control applied here is that the same control is applied to both the consistency of the bistable model, the excitable case requires
the PY and IN variables, while the TC and RE variables are not this extra step of making the stimulus more robust.
controlled. This would be desirable in the experimental scenario
where a stimulus may activate multiple neuron types with 3.4. Optimal Control of Heterogeneous
the same waveform morphology and is nonselective. Likewise, Spike-Wave Dynamics
stimuli for the TC and RE variables could be developed using the We now apply our method to a case which incorporates patient-
same framework. derived brain connectivity data. Despite idiopathic generalized
Figure 3 shows successful SWD abatement when the model is epilepsy involving widespread bilateral brain areas, it has been
placed in the bistable setting. Interestingly the same profile can argued that heterogeneity in brain connectivity may contribute
also be used in the excitable transient parameter setting since the to seizure genesis and maintenance (Taylor et al., 2013a). Indeed,
flows in state space are similar (Figure S3). it has been suggested that an improved understanding of the
heterogeneities involved may lead to more effective treatments
3.3. Optimal Control of Stochastic Spike-Wave for spike-wave seizures (Blumenfeld, 2005). We therefore
Dynamics incorporate patient-specific heterogeneous brain connectivity
The simulated seizures shown in Figure 3 are artificial in the into our model.
sense that they are induced by a stimulus at 3 s, indicated by For comparison we include a clinical recording of a
the arrow in state space. In Figure 4 we test the capability generalized SWD seizure in Figure 5A. Figure 5B shows a
of the control stimulus to abate a spontaneously occurring simulation of the model which incorporates the patient based
simulated seizure with the inclusion of noise. This has proven structural connectivity. The model is capable of reproducing
extremely challenging in a previous study where noise has various features seen clinically, specifically with respect to
been shown to impact the success rate significantly (Taylor spatial variation between recording electrodes. Three simulated
et al., 2014b). For comparison, the upper panel of Figure 4 channels are zoomed to enable closer examination. They show
high, and low amplitude spikes (first two panels) in addition to
slow wave oscillations, all of these features are routinely observed
clinically (for examples see e.g., Baier et al., 2012 and Figure 5A).
To abate the simulated seizure we apply our optimal control
method to all simulated cortical brain areas. Figure 5C shows a
time series of a simulated seizure with the control enabled. With
the exception of the controls being applied, the model parameters
and noise are identical to that shown in Figure 5B. With the
control stimuli applied the simulated seizure is terminated
almost immediately in all channels. This is despite the spatial
FIGURE 3 | Control of bistable SWD. (A) Time series of model and control heterogeneity in waveform morphology across channels and
in the bistable parameter setting (as used in Figure 2). Projection of the PY stimuli. The control signals are shown for three of the simulated
and IN variables in phase space are shown in (B). Red triangle indicates the brain areas in red in Figure 5C. There are some noticeable
trigger point at which the control was applied. The large arrow indicates the
differences in morphology and amplitude between the channels.
stimulus to induce the SWD.
For example, the bottom of the three panels has a much larger

Frontiers in Neuroscience | www.frontiersin.org 5 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

FIGURE 4 | Clinical and simulated stochastic time series with and without control. (A) Patient recording from a scalp electrode during a seizure. (B)
Stochastic model simulation without control. (C) Stochastic model simulation with control turned on.

positive deflection compared to the other two at the start, while controls developed through optimal control provide several key
at the end the negative deflection is much weaker. Due to the advantages. In contrast, much of the work in neuroscience
underlying heterogeneity some brain areas require more total using optimal control has dealt with stylized models that are
energy to abate (absolute sum of power over time). In essence analytically tractable (Moehlis et al., 2006; Li et al., 2013). Such
the total control needs to be stronger for some brain areas than analytic results provide a unique level of intuition, however, are
others. Figure 5D shows the strength of stimulus applied for not scalable to general large scale cases. We differentiate our
optimal control in different brain areas. Superior frontal areas work in this paper from the existing literature using control
(more red areas) require more power than occipital areas (more theory for neuroscience applications in the following ways.
white in color). Trigerred stimuli are applied on an “as needed” basis (i.e.,
only when the SWD reaches a trigger point) in contrast to
4. Discussion continuous feedback controllers which are always on. From a
patient perspective, this means that neurological function is
In this study we have applied optimal control to a model identical to pretreatment during the times between seizures. In
of epileptiform SWD oscillations incorporating patient-derived contrast, feedback controllers continue to operate and may as a
connectivity to prematurely abate the simulated seizure. To consequence abate non-pathological neurological activity. While
our knowledge this is the first study to incorporate diffusion non-feedback methods are often criticized for lack of robustness
MRI based connectivity from a patient into a macroscopic to noise and parameter uncertainties, recent development in
model of epilepsy and also the first attempt at simulating ensemble control allow robust open-loop controllers to be
control using a human derived DTI network. We showed that developed and demonstrated in past work with the model used
the control can work in different settings (excitable/bistable, in this paper (Ruths and Li, 2012; Ruths et al., 2014). One
stochastic/deterministic) and with different spatial properties limitation of optimal control techniques is that they are highly
(space-independent, heterogeneously spatially-extended). dependent on the ability of the model to capture the clinically
Previous modeling attempts of seizure control have included observed EEG. While this is a limitation, models for neurological
several different approaches. One approach is to apply single behavior are consistently improving, and the method for control
pulse perturbations in state space beyond the manifold which presented is highly general, so it can be applied to most models
separates the seizure and non-seizure attractor (Suffczynski et al., developed in the future. The benefit gained from a known model
2004; Taylor et al., 2014b). While there is obvious appeal to single is that the system is transferred reliably between the states
pulse stimulation, there are many difficulties with that approach, of interest (seizure state to background state). The underlying
especially in stochastic systems where repeated success can be premise of optimal control is that systems have moments in
troublesome (Taylor et al., 2014b). their dynamics when they are most and least susceptible to
A second approach leverages methods from feedback control external influence. The optimization process teases out these
theory (Kramer et al., 2006; Ching et al., 2012). While feedback susceptible periods and designs the stimulus to take advantage
control is the hallmark approach to deal with uncertainty, the of them. Although feedback control can deliver a stimulus that

Frontiers in Neuroscience | www.frontiersin.org 6 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

FIGURE 5 | Control derived using patient-specific connectivity. seizure lasts several seconds. Control is shown in red in (C) in
(A) Clinical EEG recording of a SWD seizure from 19 scalp three inset panels. (D) Spatial distribution of the total strength
electrodes. (B,C) show time series of simulated activity without and required to control the seizure. Warmer colors indicate a greater
with the control switched on. Without the control the simulated strength is applied in those areas.

adapts according to the state, it is typically sub-optimal because that all cortical variables are available for control. This may be
it has no such information about susceptibility. Optimal control undesirable experimentally as a more spatially localized solution
permits generating stimuli that are minimal by design, so that may be sought, effectively reducing the number of locations
the stimulus achieves the objective with the lowest, e.g., energy or that require stimulation to abate the seizure. While such an
duration. Finally, the stimuli found through the optimal control optimal control problem is easy to formulate, solving this mixed-
process provide intuition on the nature and dynamics of the of integer problem is challenging on a problem of this size. An
the system. important direction of our future work will seek to minimize the
There are several benefits to the control strategy used here. number of cortical areas stimulated through a variety of heuristic
First, only a subset of all variables are controlled, in this case we approaches. A further benefit is that separate controls for each
only control the cortical variables PY and IN. In the experimental variable do not necessarily need to be developed for each variable.
setting this may be desirable because external noninvasive stimuli We have demonstrated this throughout, where the same control
(e.g., transcranial magnetic stimulation) may not fully penetrate has been applied to both the PY and IN populations (see e.g.,
to deep subcortical structures such as the thalamus. In our control Figure 4). Additionally, since the control profile is precomputed,
of the spatially extended model, the control is optimal in the the delivery of the control could be applied in real-time when
sense that a cost function is optimized, given the consideration “trigger points” on the SWD cycle are detected.

Frontiers in Neuroscience | www.frontiersin.org 7 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

In this study, the same optimal controls are applied to seizure onset and offset (Jirsa et al., 2014). Indeed, our control
both PY and IN, rather than developing different controls strategy developed here could easily be applied to such a slow
for PY and for IN. In some experimental scenarios, it may variable as it would be incorporated as a state in an enlarged
be advantageous to differentiate these neuron populations, for model.
example, when using noninvasive stimuli such as TMS if the We have incorporated clinical data into our model in the form
model does not capture the variables controlled by the stimulus. of the connectivity, however, a next step is to perform the control
In other applications this may not be necessary, such as for stimuli in vivo. This could be performed first in animal models
invasive stimuli like optogenetics—where the specific variables of SWD (Meeren et al., 2005), using high strength diffusion MRI
are thought to be well known (Selvaraj et al., 2014). Furthermore, to generate high resolution connectivity matrices (Besson et al.,
the low dimensionality of SWD oscillations leads us to suggest 2014). Furthermore, with active perturbation it may be possible
that only few variables may need to be controlled (Babloyantz to elucidate the directionality of connections (David et al., 2013),
and Destexhe, 1986). Nonetheless, the method presented here which would allow for the the application of network control
is adaptable to generating either simultaneous or differentiated theory (Liu et al., 2011; Ruths and Ruths, 2014).
control signals for the various neuron populations; this choice is To summarize, we have demonstrated a nonlinear optimal
driven based on the manner in which the stimulus interacts with control technique with application to epilepsy. We have
the tissue. demonstrated its robustness in different settings, ultimately
Interestingly the total strength of control required is different building up to a large scale model of the brain which includes
in different areas (Figure 5D). Specifically the lingual gyrus, cortical connectivity derived from a patient with idiopathic
which is important for vision, required high strength bilaterally. generalized seizures. We found that due to the heterogeneity
We hypothesize this may be due to a hyperexcitability which may in connectivity, there is heterogeneity in the optimal control
exist for photoparoxysmal response, which is common in patients applied. We therefore suggest this should be considered when
with IGE and absence epilepsy as is the patient studied here. We applying stimulation to large cortical areas in vivo and that
also find superior frontal areas to require high stimulus strength. spatially localized solutions may consequently be more desirable.
Indeed, superior frontal areas are heavily involved in spike-
wave seizures with many patients exhibiting frontally abnormal Acknowledgments
activity in EEG and functional MRI recordings during seizures
(Moeller et al., 2008; Bai et al., 2010). While many IGE patients PT and MK were supported by the Engineering and
do have high amplitude abnormal frontal activity during seizures, Physical Sciences Research Council of the United Kingdom
abnormal activity in other areas is often more patient-specific. (EP/K026992/1) as part of the Human Green Brain Project
This stereotypy is present in both the spatial and temporal aspects (http://www.greenbrainproject.org). JT and JR were supported
of the seizures in many patients (Schindler et al., 2011). Indeed, by the International Design Centre (Grant IDG31300103). NS
as the seizure patterns exhibit stereotypy, even beyond SWD and JD were supported by the Ministry of Education (MOE)
seizures, so may the optimal control profiles. Tier 1 Grant M4011102 RGC3/13. PT thanks Gerold Baier and
One of the assumptions of our study is that the background Yujiang Wang for discussions regarding the model.
state coexists with the SWD limit cycle in the state space. This
is essentially a different mechanistic assumption to that of a Supplementary Material
parameter change as in some previous studies (Breakspear et al.,
2006). In that case, control of the slowly varying parameter The Supplementary Material for this article can be found
can abate the seizure. In a recent study the modulation of a online at: http://journal.frontiersin.org/article/10.3389/fnins.
parameter was implemented as an ultra-slow variable to cause 2015.00202/abstract

References Baier, G., Goodfellow, M., Taylor, P., Wang, Y., and Garry, D. (2012). The
importance of modelling epileptic seizure dynamics as spatio-temporal
Amari, S. (1977). Dynamics of pattern formation in lateral-inhibition type neural patterns. Front. Physiol. 3:281. doi: 10.3389/fphys.2012.00281
fields. Biol. Cybern. 27, 77–87. doi: 10.1007/BF00337259 Berényi, A., Belluscio, M., Mao, D., and Buzsáki, G. (2012). Closed-loop control
Asconapé, J., and Penry, J. (1984). Some clinical and eeg aspects of benign of epilepsy by transcranial electrical stimulation. Science 337, 735–737. doi:
juvenile myoclonic epilepsy. Epilepsia 25, 108–114. doi: 10.1111/j.1528- 10.1126/science.1223154
1157.1984.tb04163.x Besson, P., Lopes, R., Leclerc, X., Derambure, P., and Tyvaert, L. (2014).
Babajani-Feremi, A., and Soltanian-Zadeh, H. (2010). Multi-area neural mass Intra-subject reliability of the high-resolution whole-brain structural
modeling of EEG and MEG signals. Neuroimage 52, 793–811. doi: connectome. Neuroimage 102, 283–293. doi: 10.1016/j.neuroimage.2014.
10.1016/j.neuroimage.2010.01.034 07.064
Babloyantz, A., and Destexhe, A. (1986). Low-dimensional chaos in an instance of Blumenfeld, H. (2005). Cellular and network mechanisms of spike-wave seizures.
epilepsy. Proc. Natl. Acad. Sci. 83, 3513–3517. doi: 10.1073/pnas.83.10.3513 Epilepsia 46, 21–33. doi: 10.1111/j.1528-1167.2005.00311.x
Bai, X., Vestal, M., Berman, R., Negishi, M., Spann, M., Vega, C., et al. Blumenfeld, H., and Taylor, J. (2003). Why do seizures cause loss of consciousness?
(2010). Dynamic time course of typical childhood absence seizures: EEG, Neuroscientist 9, 301–310. doi: 10.1111/j.1528-1167.2005.00311.x
behavior, and functional magnetic resonance imaging. J. Neurosci. 30, 5884. doi: Breakspear, M., Roberts, J., Terry, J., Rodrigues, S., Mahant, N., and Robinson,
10.1523/JNEUROSCI.5101-09.2010 P. (2006). A unifying explanation of primary generalized seizures through

Frontiers in Neuroscience | www.frontiersin.org 8 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

nonlinear brain modeling and bifurcation analysis. Cereb. Cortex 16, 1296. doi: Messé, A., Rudrauf, D., Benali, H., and Marrelec, G. (2014). Relating structure
10.1093/cercor/bhj072 and function in the human brain: relative contributions of anatomy, stationary
Canuto, C., Hussaini, M., Quarteroni, A., and Zang, T. (2006). Spectral Methods: dynamics, and non-stationarities. PLoS Comput. Biol. 10:e1003530. doi:
Fundamentals in Single Domains. Berlin: Springer. 10.1371/journal.pcbi.1003530
Ching, S., Brown, E. N., and Kramer, M. A. (2012). Distributed control in a mean- Moehlis, J., Shea-Brown, E., and Rabitz, H. (2006). Optimal inputs for
field cortical network model: implications for seizure suppression. Phys. Rev. E phase models of spiking neurons. J. Comput. Nonlinear Dyn. 1, 358. doi:
86:021920. doi: 10.1103/PhysRevE.86.021920 10.1115/1.2338654
Conte, A., Gilio, F., Iacovelli, E., Bettolo, C., Di Bonaventura, C., Frasca, V., et al. Moeller, F., Siebner, H., Wolff, S., Muhle, H., Granert, O., Jansen, O., et al. (2008).
(2007). Effects of repetitive transcranial magnetic stimulation on spike-and- Simultaneous EEG-fMRI in drug-naive children with newly diagnosed absence
wave discharges. Neurosci. Res. 57, 140–142. doi: 10.1016/j.neures.2006.09.015 epilepsy. Epilepsia 49, 1510–1519. doi: 10.1111/j.1528-1167.2008.01626.x
David, O., Job, A.-S., De Palma, L., Hoffmann, D., Minotti, L., and Kahane, P. Pinault, D., and O’Brien, T. (2005). Cellular and network mechanisms of
(2013). Probabilistic functional tractography of the human cortex. Neuroimage genetically-determined absence seizures. Thalamus Relat. Syst. 3, 181. doi:
80, 307–317. doi: 10.1016/j.neuroimage.2013.05.075 10.1017/S1472928807000209
Deco, G., Ponce-Alvarez, A., Mantini, D., Romani, G. L., Hagmann, P., and Rajna, P., and Lona, C. (1989). Sensory stimulation for inhibition of
Corbetta, M. (2013). Resting-state functional connectivity emerges from epileptic seizures. Epilepsia 30, 168–174. doi: 10.1111/j.1528-1157.1989.
structurally and dynamically shaped slow linear fluctuations. J. Neurosci. 33, tb05450.x
11239–11252. doi: 10.1523/JNEUROSCI.1091-13.2013 Robinson, P. A., Rennie, C. J., and Rowe, D. L. (2002). Dynamics of large-scale
Desikan, R. S., Ségonne, F., Fischl, B., Quinn, B. T., Dickerson, B. C., Blacker, brain activity in normal arousal states and epileptic seizures. Phys. Rev. E
D., et al. (2006). An automated labeling system for subdividing the human 65:041924. doi: 10.1103/PhysRevE.65.041924
cerebral cortex on mri scans into gyral based regions of interest. Neuroimage Ruths, J., and Li, J.-S. (2012). Optimal control of inhomogeneous ensembles.
31, 968–980. doi: 10.1016/j.neuroimage.2006.01.021 Autom. Control IEEE Trans. 57, 2021–2032. doi: 10.1109/TAC.2012.
Destexhe, A. (1998). Spike-and-wave oscillations based on the properties of 2195920
GABAB receptors. J. Neurosci. 18:9099. Ruths, J., and Ruths, D. (2014). Control profiles of complex networks. Science 343,
Fornberg, B. (1998). A Practical Guide to Pseudospectral Methods, Vol. 1. 1373–1376. doi: 10.1126/science.1242063
Cambridge: Cambridge University Press. Ruths, J., Taylor, P., and Dauwels, J. (2014). “Optimal control of an epileptic neural
Haimovici, A., Tagliazucchi, E., Balenzuela, P., and Chialvo, D. R. (2013). Brain population model,” in International Federation of Automatic Control World
organization into resting state networks emerges at criticality on a model of the Congress, Vol. 19 (Cape Town), 3116–3121.
human connectome. Phys. Rev. Lett. 110:178101. doi: 10.1073/pnas.0811168106 Sadleir, L., Farrell, K., Smith, S., Connolly, M., and Scheffer, I. (2006).
Honey, C., Sporns, O., Cammoun, L., Gigandet, X., Thiran, J.-P., Meuli, R., Electroclinical features of absence seizures in childhood absence epilepsy.
et al. (2009). Predicting human resting-state functional connectivity from Neurology 67, 413–418. doi: 10.1212/01.wnl.0000228257.60184.82
structural connectivity. Proc. Natl. Acad. Sci. U.S.A. 106, 2035–2040. doi: Saillet, S., Gharbi, S., Charvet, G., Deransart, C., Guillemaud, R., Depaulis, A., et al.
10.1073/pnas.0811168106 (2012). Neural adaptation to responsive stimulation: a comparison of auditory
Jirsa, V. K., Stacey, W. C., Quilichini, P. P., Ivanov, A. I., and Bernard, C. and deep brain stimulation in a rat model of absence epilepsy. Brain Stimul. 6,
(2014). On the nature of seizure dynamics. Brain 137, 2210–2230. doi: 241–247. doi: 10.1016/j.brs.2012.05.009
10.1093/brain/awu133 Schindler, K., Gast, H., Stieglitz, L., Stibal, A., Hauf, M., Wiest, R., et al. (2011).
Kalitzin, S., Velis, D., and Lopes da Silva, F. (2010). Stimulation-based anticipation Forbidden ordinal patterns of periictal intracranial EEG indicate deterministic
and control of state transitions in the epileptic brain. Epilepsy Behav. 17, dynamics in human epileptic seizures. Epilepsia 52, 1771–1780. doi: 10.1111/j.
310–323. doi: 10.1016/j.yebeh.2009.12.023 1528-1167.2011.03202.x
Keränen, T., Sillanpää, M., and Riekkinen, P. J. (1988). Distribution of seizure Selvaraj, P., Sleigh, J. W., Freeman, W. J., Kirsch, H. E., and Szeri, A. J. (2014). Open
types in an epileptic population. Epilepsia 29, 1–7. doi: 10.1111/j.1528- loop optogenetic control of simulated cortical epileptiform activity. J. Comput.
1157.1988.tb05089.x Neurosci. 36, 515–525. doi: 10.1007/s10827-013-0484-2
Kramer, M. A., and Cash, S. S. (2012). Epilepsy as a disorder of cortical network Sporns, O., Tononi, G., and Kötter, R. (2005). The human connectome: a
organization. Neuroscientist 18, 360–372. doi: 10.1177/1073858411422754 structural description of the human brain. PLoS Comput. Biol. 1:e42. doi:
Kramer, M. A., Lopour, B. A., Kirsch, H. E., and Szeri, A. J. (2006). 10.1371/journal.pcbi.0010042
Bifurcation control of a seizing human cortex. Phys. Rev. E 73:041928. doi: Suffczynski, P., Kalitzin, S., and Lopes Da Silva, F. (2004). Dynamics of non-
10.1103/PhysRevE.73.041928 convulsive epileptic phenomena modeled by a bistable neuronal network.
Li, J.-S., Dasanayake, I., and Ruths, J. (2013). Control and synchronization Neuroscience 126, 467–484. doi: 10.1016/j.neuroscience.2004.03.014
of neuron ensembles. IEEE Trans. Autom. Control 58, 1919–1930. doi: Takens, F. (1981). “Detecting strange attractors in turbulence,” in Dynamical
10.1109/TAC.2013.2250112 Systems and Turbulence, Warwick 1980, eds R. David and Y. Lai-Sang (Berlin;
Liang, S.-F., Shaw, F.-Z., Chang, D.-W., Young, C.-P., Wang, Y. L., and Wu, S. Y. Heidelberg: Springer), 366–381.
(2012). “Live demonstration: a portable closed-loop seizure controller in freely Taylor, P., and Baier, G. (2011). A spatially extended model for macroscopic spike-
moving rats,” in Biomedical Circuits and Systems Conference (BioCAS), 2012 wave discharges. J. Comput. Neurosci. 31, 679–684. doi: 10.1007/s10827-011-
IEEE (Hsinchu: IEEE), 88–88. 0332-1
Liang, S.-F., Shaw, F.-Z., Young, C.-P., Chang, D.-W., and Liao, Y.-C. (2010). Taylor, P. N., Baier, G., Cash, S. S., Dauwels, J., Slotine, J.-J., and Wang, Y.
“A closed-loop brain computer interface for real-time seizure detection and (2013a). “A model of stimulus induced epileptic spike-wave discharges,” in IEEE
control,” in Engineering in Medicine and Biology Society (EMBC), 2010 Annual Symposium Series on Computational Intelligence (Singapore), 53–59.
International Conference of the IEEE (IEEE), 4950–4953. Taylor, P. N., Goodfellow, M., Wang, Y., and Baier, G. (2013b). Towards a large-
Liu, Y., Slotine, J., and Barabási, A. (2011). Controllability of complex networks. scale model of patient-specific epileptic spike-wave discharges. Biol. Cybern.
Nature 473, 167–173. doi: 10.1038/nature10011 107, 83–94. doi: 10.1007/s00422-012-0534-2
Luenberger, D. G. (1968). Optimization by Vector Space Methods. Hoboken, NJ: Taylor, P. N., Kaiser, M., and Dauwels, J. (2014a). Structural connectivity based
John Wiley & Sons. whole brain modelling in epilepsy. J. Neurosci. Methods 236, 51–57. doi:
Marten, F., Rodrigues, S., Suffczynski, P., Richardson, M., and Terry, J. (2009). 10.1016/j.jneumeth.2014.08.010
Derivation and analysis of an ordinary differential equation mean-field model Taylor, P. N., Wang, Y., Goodfellow, M., Dauwels, J., Moeller, F., Stephani, U., et al.
for studying clinically recorded epilepsy dynamics. Phys. Rev. E 79:021911. doi: (2014b). A computational study of stimulus driven epileptic seizure abatement.
10.1103/PhysRevE.79.021911 PLoS ONE 9:e114316. doi: 10.1371/journal.pone.0114316
Meeren, H., van Luijtelaar, G., Lopes da Silva, F., and Coenen, A. (2005). Evolving Terry, J. R., Benjamin, O., and Richardson, M. P. (2012). Seizure generation:
concepts on the pathophysiology of absence seizures: the cortical focus theory. the role of nodes and networks. Epilepsia 53, e166–e169. doi: 10.1111/j.1528-
Arch. Neurol. 62:371. doi: 10.1001/archneur.62.3.371 1167.2012.03560.x

Frontiers in Neuroscience | www.frontiersin.org 9 June 2015 | Volume 9 | Article 202

Taylor et al. Optimal control of epileptic spike-wave dynamics

Wang, Y., Goodfellow, M., Taylor, P., and Baier, G. (2012). Phase space Conflict of Interest Statement: The authors declare that the research was
approach for modeling of epileptic dynamics. Phys. Rev. E 85:061918. doi: conducted in the absence of any commercial or financial relationships that could
10.1103/PhysRevE.85.061918 be construed as a potential conflict of interest.
Westmijse, I., Ossenblok, P., Gunning, B., and Van Luijtelaar, G. (2009). Onset and
propagation of spike and slow wave discharges in human absence epilepsy: a Copyright © 2015 Taylor, Thomas, Sinha, Dauwels, Kaiser, Thesen and Ruths. This
MEG study. Epilepsia 50, 2538–2548. doi: 10.1111/j.1528-1167.2009.02162.x is an open-access article distributed under the terms of the Creative Commons
Yan, B., and Li, P. (2013). The emergence of abnormal hypersynchronization in Attribution License (CC BY). The use, distribution or reproduction in other forums
the anatomical structural network of human brain. Neuroimage 65, 34–51. doi: is permitted, provided the original author(s) or licensor are credited and that the
10.1016/j.neuroimage.2012.09.031 original publication in this journal is cited, in accordance with accepted academic
Yeh, F.-C., Wedeen, V. J., and Tseng, W.-Y. (2010). Generalized-sampling imaging. practice. No use, distribution or reproduction is permitted which does not comply
Med. Imaging IEEE Trans. 29, 1626–1635. doi: 10.1109/TMI.2010.2045126 with these terms.

Frontiers in Neuroscience | www.frontiersin.org 10 June 2015 | Volume 9 | Article 202